Apr 11, 2024 9 min read Web3 + AI Startup News

The Web3 and AI Newsletter 19

Image generated using OpenAI's DALL-E 3

Hello, everyone, and warm welcome to all newly joined subscribers!

This is the Web3 + AI newsletter, where we explore the intersection of blockchain and artificial intelligence. You may have noticed that I missed sending out a newsletter last week. The reason is that I was busy putting together the Ultimate Web3 + AI Project Directory. However, this week I will make it up to you by publishing two separate editions: the enormous VC interest in the Web3 + AI space motivated me to spin off the Web3 + AI Investment rubric into a separate recurring post. Stay tuned for it in the coming days! here

Thank you for being here! Let's dive in!

What's New in the Web3 + AI Startup World?

Fhenix & EigenLayer to Offer Next-Gen On-Chain Confidentiality

The FHE (fully homomorphic encryption) rollup Fhenix has entered an engineering partnership with the restaking protocol and data availability layer EigenLayer. They have joined forces to build a novel FHE coprocessor infrastructure for on-chain confidential computing on Ethereum.

A coprocessor serves as a companion processor designed to offload specific computational tasks from a host chain – whether it’s Ethereum, an L2 or an L3, to a designated processor living outside the scope of the host. Unlike ZK coprocessors, which are tailored for scaling computations off-chain, FHE coprocessors specialize in maintaining data confidentiality during computation, which opens up new applications that are not otherwise possible.

The FHE cryptography schemes have existed for decades, but are currently experiencing a renaissance within the crypto industry thanks to projects like Zama, Fhenix, Inco Network. Since they allow for computations, including running AI models, to be performed on encrypted data, they could empower a new era of privacy-preserving blockchain applications.

What is more, the intended FHE coprocessor will be able to offer even greater levels of security and speed since it will also use customizable crypto-economic guarantees as ensured by the ETH restaking on EigenLayer. In other words, the EigenLayer operators will be betting their own funds (ETH stake) that the execution done by the FHE rollup is valid and correct.

Ritual Allies with Celestia

The following piece of news is the quintessence of blockchain modularity. As I have written before, unlike monolithic blockchains where a consensus, an execution, and a data availability layer are all incorporated together, the modular design allows for different protocols to serve each separate layer functionality. In that particular case, users relying on Ritual as an execution layer for AI-related applications will be able to employ Celestia to satisfy their high-throughput data availability needs.

This [Celestia's] design philosophy aligns seamlessly with Ritual’s mission to bring AI on-chain with computational integrity, privacy, censorship-resistance, and decentralized incentives, with the opportunity to constantly improve via a modular architecture. We aim to foster a decentralized ecosystem where AI and crypto not only coexist but thrive together, helping usher in an entirely new wave of companies and protocols.

A blockchain's data availability layer ensures that each block's transaction data is securely stored and accessible to every node on the network. Similarly, all data produced as a result of Ritual workflows, including AI models' inputs, outputs, prompts, or proofs, could potentially be uploaded to Celestia.

Become A Web3 + AI Supporter by Subscribing with Crypto

IoTeX Teams Up with EMC, Network3

The DePIN-centric modular blockchain IoTeX has recently announced a few AI-related partnerships. First, it teamed up with the decentralized AI platform EMC to redefine the AI landscape.

The partnership between EMC and IoTeX marks an innovative era in AI and blockchain, where DeAI will be applied to IoT development in collaboration with IoTeX. The DeAI (Decentralized AI) approach, at the heart of EMC’s innovation focus, leverages the capabilities of AI within a decentralized network to provide security, privacy, and efficiency enhancements.

Then, it welcomed Network3, a project combining the functionalities of a computation DePIN and an AI-centric L2 chain with federated learning capabilities. Network3 will employ the IoTeX infrastructure and will launch the $N3 native token on its network.

Meanwhile, IoTex is looking for seed-stage DePIN projects for the second Cohort of its DePIN Surf Accelerator. Apply here to get $50K in funding.

Phala Introduces AI-Agent Contract, Partners with FLock.io

As you can surmise from this week's updates, the AI verifiability, interoperability, and coprocessors on-chain are a big topic at the moment. That is also where the following piece of news fits: the AI Coprocessor Phala Network has introduced an upgrade of its Phat Contract, called AI-Agent Contract. Building upon its previous infrastructure featuring integrations with data indexers, diverse computation templates, and trustless connection between off-chain data and smart contracts, the project moves towards deeper integration of AI.

The AI-Agent contract serves as an AI coprocessor for blockchains and is aimed at encouraging the development of modular AI agents. The confidentiality, security, and privacy aspects are also guaranteed through Phala's trusted execution environments (TEE). Learn more below.

The company also announced a partnership with federated learning platform FLock.io to democratize AI creation.

By integrating FLock’s future ML services into Phala’s AI Agent Contract, builders are able to make use of both services combined.

Together, FLock and Phala will be able to provide the revolutionary concept of LLM-fi, which aims to monetize LLMs through a process driven by community-based API aggregation.

MyShell Unveils Open-Source and Highly-Performant Model JetMoE

MyShell seems to be confidently converting from a generative AI Avatars platform to a serious research and development lab. Together with Princeton University and Massachusetts Institute of Technology (MIT), and the AI applications builder Lepton AI, MyShell just presented JetMoE.

JetMoE is an 8B-parameter open-source Large Language Model that is capable of outperforming Meta's LLaMA2-7B, but is significantly cheaper to train. It only uses public datasets for training and can be fine-tuned using consumer-grade GPU. Moreover, “JetMoE-8B only has 2.2B active parameters during inference, which drastically lowers the computational cost”.

Allora Partners with io.net, Space and Time, and Talus

After recently releasing its testnet, Upshot's self-improving AI network, Allora struck a series of interesting partnerships. First, the computation DePIN io.net reached out to Allora “to improve its computational efficiency and model accuracy in AI and machine learning development processes”.

With the capabilities provided by Allora, io.net gains the ability to bolster the security and privacy of AI and ML computations while upholding high efficiency standards, leveraging a self-improving network of ML models to optimize outcomes. Moreover, this collaboration will facilitate secure and decentralized training of AI and ML models across diverse sectors, prioritizing privacy-preserving data analysis and federated learning techniques.

Then, Allora teamed up with the decentralized data warehouse Space and Time to create a solution for providing verifiable AI training data. By delivering sub-second zero-knowledge (ZK) proofs against both on-chain and off-chain data, Space and Time secures the data provenance and authenticity when training machine intelligence.

This collaboration enables AI agents in the Allora Network to access 100TB of verifiable blockchain data from Space and Time, ensuring the models are built on a foundation of trusted, tamper-proof data.

Finally, Allora collaborated with the AI-centric chain Talus Network to build an ecosystem of self-improving AI Agents. The two companies aim at speeding up the development of AI agents across various use cases. Listen to their joint X Spaces session to learn more.

The partnership between Talus and Allora is built on a strong foundation of complementary technologies and shared vision. Allora provides the inferences and self-improvement mechanisms, while Talus provides the models and architecture for developers.

Devolved AI & cheqd to Reward Verified Data Providers

The newly-launched AI-centric Layer-1 chain Devolved AI has joined forces with trusted data markets infrastructure provider cheqd. Striving to crowdsource verified data to feed into training AI models, the two companies intend to set up a decentralized reputation system with a decentralized identity mechanism incorporated in it. It will be used to identify and reward the contributors delivering the highest-quality data.

Our ever-evolving proof-of-value protocol will ensure that community members with the best reputations wield more power in the Devolved AI ecosystem and, of course, receive rewards for their efforts!

Noteworthy Podcasts, Articles, Events

The Ultimate Web3 + AI Project Directory

I am happy to present the Ultimate Directory of Web3 + AI Projects. Ever since I have begun writing this blog, I started stumbling upon lists of projects building at the intersection of blockchain and AI. However, they all showed just a partial picture of the space, so I decided to start populating my own. I now know that gathering all Web3 + AI projects in one place is nearly an impossible mission, since new ventures appear every day. Still, I hope to have at least collected the most important ones. I also hope for you to find it useful!

Web3M X Spaces on Crypto + Real World Assets

I got invited to take part in another Web3M X Spaces, but this time we will discuss crypto and Real World Assets (RWA). Along with AI, RWAs are one of the hottest topics within the blockchain industry right now, so I am sure the session will be worth your time. What is more, I'll be accompanied by representatives of Cogito Finance, Umbrella Network, and Psi Finance! Tune in here at 5PM CEST today!

Emad Mostaque on Decentralized AI for Bankless

One of the latest Bankless episodes was once again dedicated to AI, but offered a compelling novel perspective. Emad Mostaque, founder and former CEO of one of the most prominent AI labs, Stability AI, joined the podcast to share why he has left the company to pursue decentralized AI.

Listen to the conversation below to learn why instead of focusing on the all powerful AGI, Mostaque believes we should be working to amplify human intelligence by making AI accessible to everyone.

Disclaimer: None of this should or could be considered financial advice. You should not take my words for granted, rather, do your own research (DYOR) and share your thoughts to create a fruitful discussion.